Interaction Histories: From Experience to Action and Back Again.

Abstract

We describe an enactive, situated model of interaction history based around a growing, informationally selfstructured metric space of experience that is constructed and reconstructed as the robot engages in sensorimotor interactions with objects and people in its environment.The model shows aspects of development and learning through modification of the cognitive structure that forms the basis for action selection as a result of acting in the world.We describe robotic experiments showing prediction of the path of a ball and an interaction game "peekaboo".